fscholze/uploader-recny-model-server

Spóznawanje rěče - serwer

Container(y) twarić

docker build -f docker/Dockerfile.transcription -t offline_transcription_back .
docker build -f docker/Dockerfile.phonetics     -t phonetics_back .

Přidatne dataje wobstarać

Wotpowědne rjadowaki wutworić

mkdir -p proprietary whisper fairseqdata

proprietary

Spóznawanski system wot Fraunhofer bohužel njeje zjawne přistupne.

whisper (powšitkowny)

model wobstarać a přihotować

mkdir -p tmp && cd tmp

git clone git@github.com:ggerganov/whisper.cpp.git

git checkout v1.5.4

git clone https://huggingface.co/spaces/Korla/hsb_stt_demo

git clone https://github.com/openai/whisper

cd whisper.cpp/

mkdir -p output/hsb/whisper_small

python3 ./models/convert-h5-to-ggml.py  ../hsb_stt_demo/hsb_whisper/ ../whisper/ output/hsb/whisper_small

cd ../../

model do rjadowaka kopěrować

mkdir -p whisper/hsb/whisper_small/

cp tmp/whisper.cpp/output/hsb/whisper_small/* whisper/hsb/whisper_small/

fairseqdata

cd fairseqdata

wget https://dl.fbaipublicfiles.com/mms/asr/mms1b_all.pt

whisper (europeada 2022)

mkdir -p tmp && cd tmp

git clone git@github.com:ggerganov/whisper.cpp.git

git clone https://huggingface.co/danielzoba/whisper_small_adapted_2024_06_03

git clone https://github.com/openai/whisper

cd whisper.cpp/

git checkout v1.5.4

mkdir -p output/hsb/whisper_small_europeada

mkdir -p ggml_out

cp ../whisper_small_adapted_2024_06_03/0012_whisper-base_unified_named_entities/checkpoint-2200/* ggml_out/
cp ../whisper_small_adapted_2024_06_03/0012_whisper-base_unified_named_entities/vocab.json        ggml_out/
cp ../whisper_small_adapted_2024_06_03/0012_whisper-base_unified_named_entities/added_tokens.json ggml_out/

python3 ./models/convert-h5-to-ggml.py ggml_out/ ../whisper output/hsb/whisper_small_europeada/

Container wuwjesć

Hlej dokumentacija hłownej aplikacije: tule

Licenca

Hlej dataja "LICENSE".